GPT-6 caching update cuts cached input costs by up to 90%

If your app resends the same long system prompt or document on every call, you now pay less for the repeated part. OpenAI's prompt caching update for GPT-6, announced September 22, cuts cached input token costs by up to 90%.

OpenAI lists higher cache hit rates, new diagnostics, explicit breakpoints and controls aimed at lower latency and cost. Newsroom America also reports a caching dashboard.

Caching stores the already processed start of a prompt, so repeat calls skip that work. A breakpoint marks where the reusable part ends. OpenAI's summary doesn't say how much hit rates actually go up.